|
|
LOGISTIC RECREATION: MODELLING GENITIVE ALTERNATIONS
Cathy O' Connor Boston University Friday, April 28, 3:30 PM MJH Rm 126 Sponsored by the Stanford Humanities Center/Mellon Foundation Graduate Research Program Large corpora and relatively new statistical techniques have recently enabled linguists to evaluate hypotheses about factors affecting speakers' choices within grammatical alternations (e.g. the dative, possessive, and active/passive alternations). Three factors receive primary attention in such studies: (i) NPs in pairs of alternants show differential likelihood of particular discourse/information statuses in much the same way that constituents in `non-canonical' preposing and postposing constructions do (Ward & Birner 2004). (ii) Predictable patterns of "endweight" have been shown to factor into speaker alternant choice (Arnold et al. 2000, Wasow 2002). (iii) Referent animacy has been proposed as an influential factor, particularly in the genitive alternation (Rosenbach 2005, i.a.). Yet these factors are confounded in several ways; recent studies have provided a variety of insights about how the confounds might be dealt with. One issue that complicates the disentangling of confounds in the genitive alternation is the fact that the choice between alternants involves properties of both the `possessive' modifier and the `possessed' head. To what degree might these be confounded? Bresnan et al. (in press) have shown for the dative alternation that individual verbs governing that alternation vary in the degree to which they select arguments of a particular semantic or pragmatic character. Thus these heads may contribute their own weightings to models of the dative alternation via their semantic/pragmatic selectional preferences. The genitive alternation offers a greater variety of semantic head types than some verb-governed alternations; these classes of nominal heads present a categorization and modelling challenge, but a consideration of their contributions may shed light in turn on the constraints associated with modifiers. I will take up these issues while presenting some results of a study in which 10,000 tokens of English s-genitives and of-phrases, taken from the Brown corpus, were tagged and analyzed with respect to (i), (ii), and (iii). 
|
